GPT Image models compared

    All three take up to 16 reference images and return real transparent PNGs; they differ in frame, size and prompt length.

    Try GPT Image 1.5Compare models
    Friendly modern illustration of creative team collaborating, flat editorial style, warm palette, inclusive, no text
    GPT Image 1.5
    Premium cosmetic serum bottle in frosted glass on white marble, soft three-point studio lighting, subtle water droplets, luxury product photography, shallow depth of field, centered vertical hero composition, catalog quality
    GPT Image 2
    Photoreal editorial product still life, shot on a 50mm lens at f/5.6, deep focus. A matte charcoal kraft coffee bag stands on a honed grey limestone counter beside a small brushed-steel scoop and a scatter of roasted beans. The bag carries a crisply printed two-colour label: the words "SUNBURST ROAST" in a tall condensed sans across the upper third, and beneath it in small caps "SINGLE ORIGIN / 250 g". Printing is real ink on uncoated stock, with faint fibre texture and a slightly imperfect registration on one letter edge. Hard low side light from the left through a window, warm daylight around 5200K, long soft-edged shadows falling right, one tonally even wall area in the upper right formed by the falloff of that same light. Visible material detail: paper grain, tiny dents on the steel scoop, dust on the stone, a faint oil sheen on the beans. Muted natural palette, greys and warm browns, no colour grade. No watermark, no signature, no stray text beyond the quoted label words, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no teal and orange grade, no floating particles, no centred symmetric composition.
    GPT Image 2.5

    The lineup

    Friendly modern illustration of creative team collaborating, flat editorial style, warm palette, inclusive, no text

    Prompt: Friendly modern illustration of creative team collaborating, flat editorial style, warm palette, inclusive, no text

    • 1:1
    Editorial illustration of city and technology, controlled stylized look, magazine cover energy, no text

    Prompt: Editorial illustration of city and technology, controlled stylized look, magazine cover energy, no text

    • 1:1
    SaaS analytics dashboard mockup on wide monitor, charts and KPI cards, clean light UI, no readable numbers or text

    Prompt: SaaS analytics dashboard mockup on wide monitor, charts and KPI cards, clean light UI, no readable numbers or text

    • 1:1

    GPT Image 1.5

    Three frame shapes, two quality tiers, short prompts.

    The oldest and most constrained model: three frame shapes, no resolution setting and a 3,000 character prompt ceiling. Quality is a tier, Medium or High, rather than a size.

    Best for

    • Square stickers and icons with transparency
    • Portrait or landscape stills from short briefs
    • Continuing a GPT Image 1.5 project

    Pick another model for

    • 16:9, 9:16 or 4:5 placements
    • Briefs over 3,000 characters, or print sizes

    Specs

    Inputs
    Text, optional 1 to 16 images
    Aspect ratios
    1:1, 2:3, 3:2
    Quality tiers
    Medium, High
    Resolution setting
    None
    Prompt length
    Up to 3,000 characters
    Transparent PNG
    Yes, confirmed on Medium
    Try GPT Image 1.5
    See prices

    Price

    OptionRateUnit
    Medium quality10credits
    High quality32credits
    Current rates in credits.
    Premium cosmetic serum bottle in frosted glass on white marble, soft three-point studio lighting, subtle water droplets, luxury product photography, shallow depth of field, centered vertical hero composition, catalog quality

    Prompt: Premium cosmetic serum bottle in frosted glass on white marble, soft three-point studio lighting, subtle water droplets, luxury product photography, shallow depth of field, centered vertical hero composition, catalog quality

    • 1:1
    High fashion editorial, model in structured charcoal wool blazer, dramatic side light, raw concrete wall, magazine cover energy, three quarter body, muted luxury palette

    Prompt: High fashion editorial, model in structured charcoal wool blazer, dramatic side light, raw concrete wall, magazine cover energy, three quarter body, muted luxury palette

    • 1:1
    Hero product shot, white leather sneakers on a subtle gradient plinth, soft shadow, purple to teal background, e-commerce D2C catalog, crisp studio lighting

    Prompt: Hero product shot, white leather sneakers on a subtle gradient plinth, soft shadow, purple to teal background, e-commerce D2C catalog, crisp studio lighting

    • 1:1
    Cinematic city skyline at blue hour, neon pink and warm amber light trails, key visual for a tech brand campaign, wide atmospheric composition, no readable signage

    Prompt: Cinematic city skyline at blue hour, neon pink and warm amber light trails, key visual for a tech brand campaign, wide atmospheric composition, no readable signage

    • 1:1

    GPT Image 2

    1K to 4K, six ratios, non-square above 1K.

    GPT Image 2 added 1K, 2K and 4K output and a 20,000 character prompt. At 2K and 4K it rejects square and auto before anything is charged, so a 4K square needs 2.5.

    Best for

    • Widescreen or vertical renders at 4K
    • Long, layout heavy prompts
    • Extending an existing GPT Image 2 series

    Pick another model for

    • Square or auto at 2K and 4K
    • 21:9, 3:2 or 2:3 frames

    Specs

    Inputs
    Text, optional 1 to 16 images
    Resolution
    1K, 2K, 4K
    Aspect ratios
    auto, 1:1, 9:16, 16:9, 4:3, 3:4
    2K and 4K rule
    Only 9:16, 16:9, 4:3 or 3:4
    Prompt length
    Up to 20,000 characters
    Transparent PNG
    Yes, when the prompt asks
    Try GPT Image 2
    See prices

    Price

    OptionRateUnit
    Text to image, 1K14credits
    Text to image, 2K19credits
    Text to image, 4K27credits
    Edit, 1K14credits
    Edit, 2K19credits
    Edit, 4K27credits
    Current rates in credits.
    Photoreal editorial product still life, shot on a 50mm lens at f/5.6, deep focus. A matte charcoal kraft coffee bag stands on a honed grey limestone counter beside a small brushed-steel scoop and a scatter of roasted beans. The bag carries a crisply printed two-colour label: the words "SUNBURST ROAST" in a tall condensed sans across the upper third, and beneath it in small caps "SINGLE ORIGIN / 250 g". Printing is real ink on uncoated stock, with faint fibre texture and a slightly imperfect registration on one letter edge. Hard low side light from the left through a window, warm daylight around 5200K, long soft-edged shadows falling right, one tonally even wall area in the upper right formed by the falloff of that same light. Visible material detail: paper grain, tiny dents on the steel scoop, dust on the stone, a faint oil sheen on the beans. Muted natural palette, greys and warm browns, no colour grade. No watermark, no signature, no stray text beyond the quoted label words, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no teal and orange grade, no floating particles, no centred symmetric composition.

    Prompt: Photoreal editorial product still life, shot on a 50mm lens at f/5.6, deep focus. A matte charcoal kraft coffee bag stands on a honed grey limestone counter beside a small brushed-steel scoop and a scatter of roasted beans. The bag carries a crisply printed two-colour label: the words "SUNBURST ROAST" in a tall condensed sans across the upper third, and beneath it in small caps "SINGLE ORIGIN / 250 g". Printing is real ink on uncoated stock, with faint fibre texture and a slightly imperfect registration on one letter edge. Hard low side light from the left through a window, warm daylight around 5200K, long soft-edged shadows falling right, one tonally even wall area in the upper right formed by the falloff of that same light. Visible material detail: paper grain, tiny dents on the steel scoop, dust on the stone, a faint oil sheen on the beans. Muted natural palette, greys and warm browns, no colour grade. No watermark, no signature, no stray text beyond the quoted label words, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no teal and orange grade, no floating particles, no centred symmetric composition.

    • 16:9
    • 2K
    • Sunburst
    Photoreal packaging shot, 85mm lens at f/8, deep focus, three quarter angle. A folding carton and a frosted glass dropper bottle stand on a pale plaster shelf against a warm off-white wall. The carton is printed in two inks on uncoated board: "NIGHT SERUM" in a small refined serif on the front panel, under it "30 ml / squalane and niacinamide" in fine print, and a thin rule between them. The bottle wears a matching label with the same wordmark reduced. Real print detail: crisp edges, a faint deboss along the fold, slight ink absorption into the board fibre. Light is a single soft window source from camera right at roughly 5000K, shadows falling left and softening across the plaster, a tonally even wall zone in the upper left made by that falloff. Neutral muted palette, bone, clay and deep green, no colour grade. Small honest imperfections: a fingerprint smudge on the glass, one dust speck on the shelf, an uneven shelf edge. No watermark, no signature, no stray text beyond the quoted label copy, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no heavy bokeh, no centred symmetric composition, no floating particles.

    Prompt: Photoreal packaging shot, 85mm lens at f/8, deep focus, three quarter angle. A folding carton and a frosted glass dropper bottle stand on a pale plaster shelf against a warm off-white wall. The carton is printed in two inks on uncoated board: "NIGHT SERUM" in a small refined serif on the front panel, under it "30 ml / squalane and niacinamide" in fine print, and a thin rule between them. The bottle wears a matching label with the same wordmark reduced. Real print detail: crisp edges, a faint deboss along the fold, slight ink absorption into the board fibre. Light is a single soft window source from camera right at roughly 5000K, shadows falling left and softening across the plaster, a tonally even wall zone in the upper left made by that falloff. Neutral muted palette, bone, clay and deep green, no colour grade. Small honest imperfections: a fingerprint smudge on the glass, one dust speck on the shelf, an uneven shelf edge. No watermark, no signature, no stray text beyond the quoted label copy, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no heavy bokeh, no centred symmetric composition, no floating particles.

    • 4:3
    • 1K
    • Sunburst
    Photoreal close three quarter view of a letterpress poster pinned with two brass pins to a plywood studio wall. The poster is printed on heavy cream cotton paper with a deep bite in the fibre: "MORNING LIGHT" set large in a condensed grotesque across the top, a thin horizontal rule beneath it, and in small caps at the bottom "PRINTED IN TWO INKS / NUMBER 14 OF 40". A hand-mixed rust ink and a soft slate ink, slightly uneven coverage where the plate kissed the sheet. Raking window light from the left at about 4800K makes the deboss of each letter cast its own micro shadow; the paper edge is deckled and curls a few millimetres away from the wall on the lower right. Plywood grain, a couple of old pinholes, a faint pencil mark near the corner. Muted natural palette, no colour grade, deep focus, 50mm lens. No watermark, no signature, no stray text beyond the quoted poster copy, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no heavy bokeh, no floating particles.

    Prompt: Photoreal close three quarter view of a letterpress poster pinned with two brass pins to a plywood studio wall. The poster is printed on heavy cream cotton paper with a deep bite in the fibre: "MORNING LIGHT" set large in a condensed grotesque across the top, a thin horizontal rule beneath it, and in small caps at the bottom "PRINTED IN TWO INKS / NUMBER 14 OF 40". A hand-mixed rust ink and a soft slate ink, slightly uneven coverage where the plate kissed the sheet. Raking window light from the left at about 4800K makes the deboss of each letter cast its own micro shadow; the paper edge is deckled and curls a few millimetres away from the wall on the lower right. Plywood grain, a couple of old pinholes, a faint pencil mark near the corner. Muted natural palette, no colour grade, deep focus, 50mm lens. No watermark, no signature, no stray text beyond the quoted poster copy, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no heavy bokeh, no floating particles.

    • 4:3
    • 1K
    • Flare
    Photoreal close three quarter view of a letterpress poster pinned with two brass pins to a plywood studio wall. The poster is printed on heavy cream cotton paper with a deep bite in the fibre: "MORNING LIGHT" set large in a condensed grotesque across the top, a thin horizontal rule beneath it, and in small caps at the bottom "PRINTED IN TWO INKS / NUMBER 14 OF 40". A hand-mixed rust ink and a soft slate ink, slightly uneven coverage where the plate kissed the sheet. Raking window light from the left at about 4800K makes the deboss of each letter cast its own micro shadow; the paper edge is deckled and curls a few millimetres away from the wall on the lower right. Plywood grain, a couple of old pinholes, a faint pencil mark near the corner. Muted natural palette, no colour grade, deep focus, 50mm lens. No watermark, no signature, no stray text beyond the quoted poster copy, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no heavy bokeh, no floating particles.

    Prompt: Photoreal close three quarter view of a letterpress poster pinned with two brass pins to a plywood studio wall. The poster is printed on heavy cream cotton paper with a deep bite in the fibre: "MORNING LIGHT" set large in a condensed grotesque across the top, a thin horizontal rule beneath it, and in small caps at the bottom "PRINTED IN TWO INKS / NUMBER 14 OF 40". A hand-mixed rust ink and a soft slate ink, slightly uneven coverage where the plate kissed the sheet. Raking window light from the left at about 4800K makes the deboss of each letter cast its own micro shadow; the paper edge is deckled and curls a few millimetres away from the wall on the lower right. Plywood grain, a couple of old pinholes, a faint pencil mark near the corner. Muted natural palette, no colour grade, deep focus, 50mm lens. No watermark, no signature, no stray text beyond the quoted poster copy, no invented brand marks, no gibberish lettering, no mirrored text, no HDR glow, no bloom, no heavy bokeh, no floating particles.

    • 4:3
    • 1K
    • Sunburst

    GPT Image 2.5

    All 13 ratios at every size, in two lanes.

    The current model keeps 4K output and 20,000 character prompts and drops the ratio rule: all 13 ratios work at every resolution. It ships as two lanes, Sunburst for final renders and Flare for iteration and batches.

    Best for

    • Final frames with legible copy
    • Ultrawide 21:9 and tall 16:27 formats
    • Square output at 2K and 4K

    Pick another model for

    • 4:5 placements, in either lane

    Specs

    Lanes
    Flare, Sunburst
    Inputs
    Text, optional 1 to 16 images
    Resolution
    1K, 2K, 4K
    Aspect ratios
    auto, 1:1, 3:2/2:3, 4:3/3:4, 9:8/8:9
    Wide and tall ratios
    16:9/9:16, 21:9, 27:16/16:27
    Prompt length
    Up to 20,000 characters
    Try GPT Image 2.5
    See prices

    Price

    OptionRateUnit
    Flare, text to image, 1K15credits
    Flare, text to image, 2K20credits
    Flare, text to image, 4K28credits
    Flare, edit, 1K15credits
    Flare, edit, 2K20credits
    Flare, edit, 4K28credits
    Sunburst, text to image, 1K15credits
    Sunburst, text to image, 2K20credits
    Sunburst, text to image, 4K28credits
    Sunburst, edit, 1K15credits
    Sunburst, edit, 2K20credits
    Sunburst, edit, 4K28credits
    Current rates in credits.

    The three side by side

    ModelAspect ratiosResolutionSquare at 4KPrompt lengthOptions
    GPT Image 1.53No settingNo3,000Medium or High tier
    GPT Image 261K, 2K, 4KNo20,000One lane
    GPT Image 2.5131K, 2K, 4KYes20,000Flare or Sunburst lane

    Which GPT Image should I use?

    How the family fits together

    Three generations of one idea: a model that reads a long, specific brief plus reference images, then generates or edits against both. Each step widened the frame: GPT Image 2 added 1K to 4K output, and 2.5 opened 13 ratios at every size.

    Working with GPT Image

    • Quote every word to render, exactly as it should appear, so the model prints it literally.
    • Attaching an image switches any GPT Image model to editing. There is no edit toggle.
    • Need 4:5? No GPT Image model renders it: generate at 3:4 or 1:1 and crop.

    Pick a plan

    One credit balance covers images, video and audio. Pick the size that fits how much you make.

    Most popular

    Studio Pro

    Studio value with extra headroom.

    49 EURmonth
    6,400credits / month
    Team seats

    2 seats included

    What you'll generate

    • Images~1,600
    • Songs~492
    • Videos~640
    • Separations~1,066
    • Caption Minutes~213
    • All modelsIncluded
    Select plan

    Starter

    Low tier generations for regular use.

    10 EURmonth
    1,100credits / month

    What you'll generate

    • Images~275
    • Songs~84
    • Videos~110
    • Separations~183
    • Caption Minutes~36
    • All modelsIncluded
    Select plan

    Pro

    Access full experience.

    20 EURmonth
    2,600credits / month

    What you'll generate

    • Images~650
    • Songs~200
    • Videos~260
    • Separations~433
    • Caption Minutes~86
    • All modelsIncluded
    Select plan

    Powerhouse

    Most cost efficient.

    99 EURmonth
    13,000credits / month

    13,000 credits/mo

    Team seats
    3 seats

    Extra seats €19/seat/mo

    What you'll generate

    • Images~3,250
    • Songs~1,000
    • Videos~1,300
    • Separations~2,166
    • Caption Minutes~433
    • All modelsIncluded
    Select plan

    Enterprise

    For teams that outgrow standard plans.

    Custom

    Pricing tailored to your team

    What's included

    • Custom number of seats
    • Custom credit volume
    • Invoice billing
    • Priority support
    Contact sales
    Or pay-as-you-go

    Test any model with one-time credit packs

    Credits never expire and don't require a subscription — buy a pack, generate, see if Kubeez fits.

    Quick refill

    Quick refill for short bursts of generation.

    Credits

    500

    You pay

    €6

    Standard pack

    Mid-size pack at the standard per-credit rate.

    Credits

    1,500

    You pay

    €18

    −5%

    Volume pack

    Larger pack with a volume discount built in.

    Credits

    5,000

    You pay

    €60€57

    −10%

    Production pack

    Best per-credit price for sustained production.

    Credits

    15,000

    You pay

    €180€162

    Frequently asked questions

    • GPT Image 2.5: the newest, with the most ratios and 4K in all of them. Use Sunburst for the frame you ship and Flare while exploring.

    • Yes. GPT Image 1.5 Medium, GPT Image 2 and both 2.5 lanes return a real alpha channel when the prompt asks. Check the downloaded file; gallery thumbnails are drawn on white.

    • Up to 16 on all three models. A seventeenth image is rejected before the run is charged.

    • Two lanes of GPT Image 2.5 with identical inputs, ratios and sizes, picked on one studio card. Over the API they are gpt-image-2-5-flare and gpt-image-2-5-sunburst.

    • Yes. Each runs under its own id in the studio, the REST API and the MCP server, for briefs that fit them and projects started on them.

    Open GPT Image in the studio

    Pick the model that fits your job, write your prompt and run it.

    Open the studioAll model families